from lib import *

import warnings
warnings.filterwarnings('ignore')

cmap= newcmp

vmin_corr=-0.5
vmax_corr = 0.5

cutoff = 1e-15
tol = 1e-15
mineigenval, maxeigenval = 1e-20, 1e3

n_of_weights=5
n_of_biases = 5
n_of_samples = 5

Get all on one plot

def get_eigrnvals_for_plot(path,cutoff):

    eigenvals_tab = []
    
    for n in range(n_of_samples):       
        file = path + str(n+1) + '.csv'
        pearson_corr = pd.read_csv(file)

        rk = np.linalg.matrix_rank(pearson_corr, tol = tol) 
        eigenvals = np.linalg.eig(pearson_corr)[0].real    
        eigenvals = eigenvals[:cutoff]
        
        #eigenvals = np.where(eigenvals < cutoff, 1e-18, eigenvals) 
        
        eigenvals_tab.append(eigenvals)
        print("{:} matrix shape = {:}\t rank = {:}".format(path, pearson_corr.shape, rk)) 
        
    
    eigenvals_arr = np.array(eigenvals_tab)
    log_eigenvals_arr = np.log10(eigenvals_arr)
    
    eigenvals_mean = eigenvals_arr.mean(axis=0)
    eigenvals_std = eigenvals_arr.std(axis=0)
    
    log_eigenvals_mean = log_eigenvals_arr.mean(axis=0)
    log_eigenvals_std = log_eigenvals_arr.std(axis=0)
    
    return eigenvals_mean, eigenvals_std, log_eigenvals_mean

def show_correlation(folder, pwd, details, n_of_samples, cmap, vmin, vmax, mineigenval, maxeigenval, cutoff):
    
    colors = ['slategray', 'blueviolet', 'indigo','blueviolet','pink'] ##violet

    fig, axs = plt.subplots(1, 2, figsize = (15, 6))
    fig.suptitle(folder + pwd, fontsize=20)
    
    
    for i,d in enumerate(details):
        
        path = folder + pwd + '_' + d + '_'
        eigenvals_mean, eigenvals_std, log_eigenvals_mean = get_eigrnvals_for_plot(path, cutoff)
    
        m=len(eigenvals_mean)
        t = np.linspace(0, m-1, m)       
        bins = int(np.sqrt(m))  
        
        axs[0].hist(log_eigenvals_mean, bins=bins, density=True, alpha = 0.5 , color=colors[i], label=d)
        axs[0].set_title("log eigenvalues histogram")
        axs[0].grid(True, color = 'gray', alpha =0.4)
        axs[0].legend()

        axs[1].errorbar(t, eigenvals_mean, yerr=eigenvals_std, c=colors[i], alpha = 0.9, label = d)
        #axs[2].scatter(t, eigenvals_mean, color  ='indigo', s=5)

        axs[1].set_title("eigenvalues")   
        axs[1].set_xlim(0, m-1)
        axs[1].set_yscale('log')
        axs[1].set_ylim((mineigenval, maxeigenval))
        axs[1].grid(True, color = 'gray', alpha =0.4)
        axs[1].legend()
        
    

2 layer

pwd = 'model_50_50/model_50_50'
folder = 'matrix_crosscorrelation_small/'
details = [ 'w3', 'w2']
show_correlation(folder, pwd, details, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1000)

matrix_crosscorrelation_small/model_50_50/model_50_50_w3_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation_small/model_50_50/model_50_50_w3_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation_small/model_50_50/model_50_50_w3_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation_small/model_50_50/model_50_50_w3_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation_small/model_50_50/model_50_50_w3_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation_small/model_50_50/model_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2468
matrix_crosscorrelation_small/model_50_50/model_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2468
matrix_crosscorrelation_small/model_50_50/model_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2469
matrix_crosscorrelation_small/model_50_50/model_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2468
matrix_crosscorrelation_small/model_50_50/model_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2469

3 layer

pwd = 'model_50_50_50/model_50_50_50'
folder = 'matrix_crosscorrelation/'
details = ['w4', 'w3', 'w2']
show_correlation(folder, pwd, details, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1500)

matrix_crosscorrelation/model_50_50_50/model_50_50_50_w4_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w4_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w4_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w4_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w4_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2479
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2479
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2480
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2479
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2478
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2479
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2480
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2479
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2479
matrix_crosscorrelation/model_50_50_50/model_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2481

pwd = 'model_50_50_50/model_50_50_50'
folder = 'matrix_crosscorrelation_small/'
details = ['w4', 'w3', 'w2']
show_correlation(folder, pwd, details, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1500)

matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w4_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w4_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w4_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w4_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w4_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2467
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2468
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2467
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2466
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2468
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2467
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2468
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2468
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2466
matrix_crosscorrelation_small/model_50_50_50/model_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2469

4 layer

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation/'
details = ['w5', 'w4', 'w3', 'w2']
show_correlation(folder, pwd, details, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 2000)

matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w5_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w5_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w5_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w5_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w5_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w4_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w4_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w4_ matrix shape = (2500, 2500)	 rank = 2487
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w4_ matrix shape = (2500, 2500)	 rank = 2489
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w4_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2486
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2487
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2487
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2487
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2487
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2486
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2488

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation/'
details = ['w5', 'w4', 'w3', 'w2']
show_correlation(folder, pwd, details, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 2000)

matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w5_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w5_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w5_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w5_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w5_ matrix shape = (500, 500)	 rank = 500
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w4_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w4_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w4_ matrix shape = (2500, 2500)	 rank = 2487
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w4_ matrix shape = (2500, 2500)	 rank = 2489
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w4_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2486
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2487
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2487
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w3_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2487
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2488
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2487
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2486
matrix_crosscorrelation/model_50_50_50_50/model_50_50_50_50_w2_ matrix shape = (2500, 2500)	 rank = 2488

More details:

def show_correlation(folder, pwd, d, n_of_samples, cmap, vmin, vmax, mineigenval, maxeigenval, cutoff = 1e-15):

    eigenvals_tab = []
    
    for n in range(n_of_samples):       
        path = folder + pwd + '_' + d + '_'+ str(n+1) + '.csv'
        pearson_corr = pd.read_csv(path)

        rk = np.linalg.matrix_rank(pearson_corr, tol = cutoff) 
        eigenvals = np.linalg.eig(pearson_corr)[0].real    
        eigenvals_tab.append(eigenvals)
        print("matrix shape = {:}\t\t rank = {:}".format(pearson_corr.shape, rk)) 
        
    
    eigenvals_arr = np.array(eigenvals_tab)
    log_eigenvals_arr = np.log10(eigenvals_arr)
    
    eigenvals_mean = eigenvals_arr.mean(axis=0)
    eigenvals_std = eigenvals_arr.std(axis=0)
    
    log_eigenvals_mean = log_eigenvals_arr.mean(axis=0)
    log_eigenvals_std = log_eigenvals_arr.std(axis=0)
    
    
    m=len(eigenvals_mean)
    t = np.linspace(0, m-1, m)       
    bins = int(np.sqrt(m))  
        
    #eigenvals = np.where(eigenvals < cutoff, 1e-18, eigenvals) 
        
        
    fig, axs = plt.subplots(1, 3, figsize = (15, 6))
    
    fig.suptitle(folder + pwd, fontsize=20)
    
    im = axs[0].imshow(pearson_corr, cmap = cmap, vmin=vmin, vmax=vmax)
    axs[0].figure.colorbar(im, ax=axs[0])  
    axs[1].hist(log_eigenvals_mean, bins=bins, density=True, alpha = 1, color='indigo')
    axs[1].set_title("log eigenvalues histogram")
    axs[1].grid(True, color = 'gray', alpha =0.4)
    
    axs[2].errorbar(t, eigenvals_mean, yerr=eigenvals_std, c='pink', alpha = 0.5)
    axs[2].scatter(t, eigenvals_mean, color  ='indigo', s=5)

    axs[2].set_title("eigenvalues")   
    axs[2].set_xlim(0, m-1)
    axs[2].set_yscale('log')
    axs[2].set_ylim((mineigenval, maxeigenval))
    axs[2].grid(True, color = 'gray', alpha =0.4)
    

4 hidden layers

Regular size: 2000 epochs

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation/'
d='w2'

show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (2500, 2500)		 rank = 2487
matrix shape = (2500, 2500)		 rank = 2488
matrix shape = (2500, 2500)		 rank = 2487
matrix shape = (2500, 2500)		 rank = 2486
matrix shape = (2500, 2500)		 rank = 2488

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation/'
d='w3'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (2500, 2500)		 rank = 2488
matrix shape = (2500, 2500)		 rank = 2486
matrix shape = (2500, 2500)		 rank = 2487
matrix shape = (2500, 2500)		 rank = 2487
matrix shape = (2500, 2500)		 rank = 2488

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation/'
d='w4'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (2500, 2500)		 rank = 2488
matrix shape = (2500, 2500)		 rank = 2488
matrix shape = (2500, 2500)		 rank = 2487
matrix shape = (2500, 2500)		 rank = 2489
matrix shape = (2500, 2500)		 rank = 2488

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation/'
d='w5'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500

Small size: 1000 epochs

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation_small/'
d='w2'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (2500, 2500)		 rank = 2465
matrix shape = (2500, 2500)		 rank = 2464
matrix shape = (2500, 2500)		 rank = 2465
matrix shape = (2500, 2500)		 rank = 2464
matrix shape = (2500, 2500)		 rank = 2465

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation_small/'
d='w3'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (2500, 2500)		 rank = 2467
matrix shape = (2500, 2500)		 rank = 2466
matrix shape = (2500, 2500)		 rank = 2466
matrix shape = (2500, 2500)		 rank = 2466
matrix shape = (2500, 2500)		 rank = 2466

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation_small/'
d='w4'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (2500, 2500)		 rank = 2467
matrix shape = (2500, 2500)		 rank = 2467
matrix shape = (2500, 2500)		 rank = 2466
matrix shape = (2500, 2500)		 rank = 2467
matrix shape = (2500, 2500)		 rank = 2467

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation_small/'
d='w5'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500

Very small size: 500 epochs

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation_vsmall/'
d='w2'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (2500, 2500)		 rank = 2441
matrix shape = (2500, 2500)		 rank = 2441
matrix shape = (2500, 2500)		 rank = 2442
matrix shape = (2500, 2500)		 rank = 2441
matrix shape = (2500, 2500)		 rank = 2440

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation_vsmall/'
d='w3'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (2500, 2500)		 rank = 2443
matrix shape = (2500, 2500)		 rank = 2440
matrix shape = (2500, 2500)		 rank = 2443
matrix shape = (2500, 2500)		 rank = 2441
matrix shape = (2500, 2500)		 rank = 2441

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation_vsmall/'
d='w4'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (2500, 2500)		 rank = 2445
matrix shape = (2500, 2500)		 rank = 2445
matrix shape = (2500, 2500)		 rank = 2445
matrix shape = (2500, 2500)		 rank = 2444
matrix shape = (2500, 2500)		 rank = 2446

pwd = 'model_50_50_50_50/model_50_50_50_50'
folder = 'matrix_crosscorrelation_vsmall/'
d='w5'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (500, 500)		 rank = 499
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 499
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 499

3 hidden layers

Regular size: 1500 epochs

pwd = 'model_50_50_50/model_50_50_50'
folder = 'matrix_crosscorrelation/'
d='w2'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff)

matrix shape = (2500, 2500)		 rank = 2479
matrix shape = (2500, 2500)		 rank = 2480
matrix shape = (2500, 2500)		 rank = 2479
matrix shape = (2500, 2500)		 rank = 2479
matrix shape = (2500, 2500)		 rank = 2481

pwd = 'model_50_50_50/model_50_50_50'
folder = 'matrix_crosscorrelation/'
d='w3'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff)

matrix shape = (2500, 2500)		 rank = 2479
matrix shape = (2500, 2500)		 rank = 2479
matrix shape = (2500, 2500)		 rank = 2480
matrix shape = (2500, 2500)		 rank = 2479
matrix shape = (2500, 2500)		 rank = 2478

pwd = 'model_50_50_50/model_50_50_50'
folder = 'matrix_crosscorrelation/'
d='w4'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500

Small size: 1000 epochs

pwd = 'model_50_50_50/model_50_50_50'
folder = 'matrix_crosscorrelation_small/'
d='w2'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff)

matrix shape = (2500, 2500)		 rank = 2467
matrix shape = (2500, 2500)		 rank = 2468
matrix shape = (2500, 2500)		 rank = 2468
matrix shape = (2500, 2500)		 rank = 2466
matrix shape = (2500, 2500)		 rank = 2469

pwd = 'model_50_50_50/model_50_50_50'
folder = 'matrix_crosscorrelation_small/'
d='w3'
show_correlation(folder, pwd, d, n_of_samples, cmap,  vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff)

matrix shape = (2500, 2500)		 rank = 2467
matrix shape = (2500, 2500)		 rank = 2468
matrix shape = (2500, 2500)		 rank = 2467
matrix shape = (2500, 2500)		 rank = 2466
matrix shape = (2500, 2500)		 rank = 2468

pwd = 'model_50_50_50/model_50_50_50'
folder = 'matrix_crosscorrelation_small/'
d='w4'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff)

matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500

Very small size: 500 epochs

pwd = 'model_50_50_50/model_50_50_50'
folder = 'matrix_crosscorrelation_vsmall/'
d='w2'
show_correlation(folder, pwd, d, n_of_samples, cmap,  vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff)

matrix shape = (2500, 2500)		 rank = 2445
matrix shape = (2500, 2500)		 rank = 2444
matrix shape = (2500, 2500)		 rank = 2444
matrix shape = (2500, 2500)		 rank = 2443
matrix shape = (2500, 2500)		 rank = 2445

pwd = 'model_50_50_50/model_50_50_50'
folder = 'matrix_crosscorrelation_vsmall/'
d='w3'
show_correlation(folder, pwd, d, n_of_samples, cmap,  vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff)

matrix shape = (2500, 2500)		 rank = 2447
matrix shape = (2500, 2500)		 rank = 2447
matrix shape = (2500, 2500)		 rank = 2446
matrix shape = (2500, 2500)		 rank = 2444
matrix shape = (2500, 2500)		 rank = 2447

pwd = 'model_50_50_50/model_50_50_50'
folder = 'matrix_crosscorrelation_vsmall/'
d='w4'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff)

matrix shape = (500, 500)		 rank = 498
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 499
matrix shape = (500, 500)		 rank = 499
matrix shape = (500, 500)		 rank = 499

2 hidden layers

Small/ Regular size: 1000 epochs

pwd = 'model_50_50/model_50_50'
folder = 'matrix_crosscorrelation_small/'
d='w2'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (2500, 2500)		 rank = 2468
matrix shape = (2500, 2500)		 rank = 2468
matrix shape = (2500, 2500)		 rank = 2469
matrix shape = (2500, 2500)		 rank = 2468
matrix shape = (2500, 2500)		 rank = 2469

pwd = 'model_50_50/model_50_50'
folder = 'matrix_crosscorrelation_small/'
d='w3'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500
matrix shape = (500, 500)		 rank = 500

Very small size: 500 epochs

pwd = 'model_50_50/model_50_50'
folder = 'matrix_crosscorrelation_vsmall/'
d='w2'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (2500, 2500)		 rank = 2448
matrix shape = (2500, 2500)		 rank = 2449
matrix shape = (2500, 2500)		 rank = 2451
matrix shape = (2500, 2500)		 rank = 2448
matrix shape = (2500, 2500)		 rank = 2447

pwd = 'model_50_50/model_50_50'
folder = 'matrix_crosscorrelation_vsmall/'
d='w3'
show_correlation(folder, pwd, d, n_of_samples, cmap, vmin_corr, vmax_corr, mineigenval, maxeigenval, cutoff = 1e-15)

matrix shape = (500, 500)		 rank = 498
matrix shape = (500, 500)		 rank = 498
matrix shape = (500, 500)		 rank = 498
matrix shape = (500, 500)		 rank = 497
matrix shape = (500, 500)		 rank = 497